JP5487424B2 - Information embedding method and reading method in binary image - Google Patents

Description

The present invention relates to information embedding and reading in an image using a binary image having a simple shape such as a pattern or logo that can be read by a mobile phone.

  The most popular QR code at present is a black and white matrix that cannot change its shape, and the code itself has no meaning as an image, so it lacks originality and design and requires extra space. Therefore, the use of a color code, which is a two-dimensional code that can be used in several colors and can be designed to some extent, has also been seen. The color code arranges four colors in a 5 × 5 matrix, reads a number from the pattern, and acquires a URL registered in advance in the server. Although it is better in design than QR code, it is not always possible to obtain a satisfactory design because the colors that can be used and the positions of the colors are determined.

  On the other hand, a system that can use the design as it is has been developed, but this is an image search system that searches an image registered in a server, and does not itself become a code. Therefore, there arises a problem that similar images cannot be registered and a high communication cost of transmitting the images themselves. Also, FPcode, in which photographs are coded almost as they are, has been developed. This is coded by changing the yellow value, which is inconspicuous for human eyes. Although it is useful because it can be used as it is, there are drawbacks in that it is difficult to code a simple form with few colors, and it cannot be used on print media with limited colors.

  Furthermore, there is a method of scanning an image in units of pixels, determining whether to embed information according to a pixel pattern in a certain size block, and embedding by changing the pattern. Several methods have been proposed for selecting the embeddable pattern (Non-Patent Document 1). In this method, the appearance frequency of a pattern in a sentence image is examined, and a pattern that is frequently changed and inconspicuous is selected. Min et al. Have proposed a method of selecting a pattern that does not stand out even if it is changed by calculating the degree of variability from smoothness and connectivity (Non-patent Document 2). This method is not limited to text, and may be a single character or pattern, and can be applied to any binary image. However, since an inconspicuous pattern is selected, it is necessarily embedded in the edge portion.

On the other hand, Patent Document 1 is a watermark technique for preventing falsification. Although it is disclosed that the image is easily noticeable when the image is changed, a method for specifying tampering is disclosed, and in order to specify this tampering, the pixel position can be determined with an accuracy that can be determined. Is what you need.

JP-A-2004-289978

Satoshi Abe, Koichi Inoue, Koichi Ejiri, “Digital Watermarking on Binary Images”, Symposium on Cryptography and Information Security, (SCIS) 2000C05, 2000 Min Wu, Bede Liu, “Data Hiding in Binary Image for Authentication and Annotation”, IEEE TRANSACTIONS ON MULTITIMEDIA, vol. 6, no. 4, pp. 528-538, 20044.8

An object of the present invention is to embed information without destroying the shape of an original image.
The object is to detect embedded data without requiring original image information. Another object of the present invention is to realize sufficient detection capability even from acquisition at a resolution different from the resolution at the time of printing, for example, from a photographed image with a camera-equipped mobile phone. Furthermore, it is an object of the present invention to acquire embedded data without any problems even when an image is slanted or slightly distorted during photographing with a mobile phone.

In order to achieve the above object, the invention of claim 1 is based on the smoothness and connectivity of the block unit when the entire image is scanned in a predetermined block unit, and the brightness of the central pixel of the block unit is inverted. When the degree of variability of the block is calculated and the degree of variability is equal to or higher than a predetermined score, the block is identified as an information embedding block, an embedding information bit string composed of a central pixel string of the embedding block is created, and the information embedding bit string Compare the brightness and darkness with the information bit string to be embedded, and for the bits that do not match, invert the information embedding bit to embed binary information in the image, provide a predetermined padding outside the image, A direction detection space comprising predetermined pixels for direction detection is arranged at four corners, and one of the direction detection spaces is A binary image information embedding method, characterized in that darkness is embedded as a direction detection mark only in an upper left direction detection space, and dot frames are arranged outside the putting so as to alternate between light and dark, It is a summary.
As a result, it is possible to embed information without destroying the shape of the image in the binary image.

According to a second aspect of the present invention, there is provided a method of reading information embedded in a binarized image that reads out information from an image in which information is embedded by the information embedding method of the binarized image according to the first aspect. Brightness is detected from the upper and lower centers of the captured image with a width larger than the pixel width of the original image from the left end of the image toward the center, and the first detected dark position is determined in the left frame of the image. The brightness of the image is detected from the right end toward the center of the image, the dark position detected first is defined as one point on the right frame of the image, and the pixel width of the original image is moved upward from the left point. The brightness is detected with a larger width, the last detected dark position is set as the upper left corner of the frame of the image, and the left and right sides of the image are shifted from one point to the lower side with a width larger than the pixel width of the original image. Detect and last The detected dark position is set as the lower left corner of the frame of the image, and similarly, the upper and lower corners of the frame of the image are detected by detecting light and dark with a width larger than the pixel width of the original image from one point on the right frame. And the lower right corner, light and dark are detected in the direction from the upper left corner to the upper right corner, and XY coordinate values changing from light to dark and from dark to light are recorded, and from the lower left corner to the right Light and dark are detected in the direction of the lower corner, XY coordinate values that change from light to dark and dark to light are recorded, light and dark are detected from the upper left corner to the lower left corner, and light to dark and dark to dark. XY coordinate values that change brightly are recorded, light and darkness are detected in the direction from the upper right corner to the lower right corner, and XY coordinate values that change from light to dark and from dark to light are recorded, and the X coordinate values are the same. A matrix is created by connecting those having the same Y-coordinate and configuring the matrix The original image is restored with a unit of one pixel, the restored original image is scanned in a predetermined block unit, and the embedded information bit string composed of the central pixel string of the block is set as binary information embedded in the original image. The gist of the present invention is a method for reading two-image embedded information.
Thus, the information embedded by the method of claim 1 can be detected without requiring the original image information.

Furthermore, the invention described in claim 3 is the binary image information embedding comprising the binary image information embedding method according to claim 1 and the binary image embedding information reading method according to claim 2. Summary of the method and information reading method.
The code created by the method of the present invention is called Be-code.

 The use of a binarized image has the advantage that it is not affected even if the number of colors used is small and can be printed on any print medium, so that the form of use is not limited. In addition, since a simple shape image can be used, a company logo or the like can be coded.

It is a figure explaining the dot frame of Be-code. It is a flowchart which shows the procedure of Be-code preparation. It is a flowchart which shows the outline | summary of a Be-code reading procedure. It is a flowchart which shows the detection procedure of the angle | corner of the flame | frame at the time of reading. It is a flowchart which shows the analysis procedure of the flame | frame at the time of reading. It is a flowchart which shows the restoration procedure of Be-code. It is a flowchart which shows a bit information read-out procedure. Example of a system using Be-code Illustration showing smoothness It is a flowchart which investigates connectivity. It is the figure which illustrated the variable degree for determining an embedding position.

(Configuration of Be-code)
First, the configuration of the Be-code will be described with reference to FIG.
Around the Be-code 10, dot frames 110 arranged alternately in black and white for determining the unit of pixels, padding 100 where the dot frames 110 are gaps provided so as not to contact the emblem 120, Be− A direction detection mark 131 that is a mark for determining the top, bottom, left, and right of the code 10, a direction detection space 130 having a side of a predetermined length from the dot frame 110 to reliably read the direction detection mark 131, and the Be-code 10 The margin 140 is provided so as not to come into contact with other patterns or characters on the surface to be installed or printed.

  In the following, the generic name of the frame configured around the Be-code 10 is referred to as a frame. By the way, the margin 140 is not actually drawn, which means that the Be-code 10 of the present invention is drawn on a medium such as a poster or a flyer apart from other patterns and characters.

(Be-code creation method)
The creation flow of the Be-code 10 will be described with reference to FIG.
As shown in FIG. 2, an image to be an emblem 120 is obtained through the steps of image binarization (S 110), bit information embedding (S 120), frame addition (S 130), and colorization (S 140). The Be-code 10 is generated from the bit information to be embedded.

(Input of original image and embedded data)
The original image is preferably a simple image such as a pattern or logo. Even illustrations with gradations that cross light and dark and natural images with a wide luminance range can be coded, but this is not very suitable because of the large change in appearance. As preprocessing for encoding, an operation of changing the size of the input image is performed so that darkness is arranged at the four corners of the dot frame. In the present invention, when the number of vertical or horizontal pixels of the input image is an even number, the end column or row is cut by one pixel.
The embedded data content is arbitrary in this paper.

(Binarization of input image)
S110 will be described. The input image is binarized. In the present invention, binarization is performed using the median value of the maximum luminance and the minimum luminance in the original image as threshold values. This is to prevent the image from becoming an inappropriate binary image when performing simple binarization by considering the brightness of the entire image, and many other methods have been proposed. It is not fixed to this.

(Embedding information)
S120 will be described. The encoded bit string is embedded in the binarized image. For embedding in a binary image, there is no choice but to invert light to dark and dark to light, and selection of the part to be inverted is important. Therefore, in the present invention, a digital watermark to a binary image using smoothness and connectivity proposed by the Min method shown in Non-Patent Document 2 is used as a basis.

In this method, the input image is referenced for each small area (block), and the degree of variability is calculated from the pixel group in each block using smoothness and connectivity. The degree of variability is a score with little change in appearance without losing smoothness connectivity even when the darkness of the central pixel in the block is reversed. The block size is 3 × 3 pixels centered on the target pixel. If the block size is increased to 5 × 5, the appearance change is not impaired, but the amount of information that can be embedded is significantly reduced.

Smoothness has four directions, vertical, horizontal, diagonal, and reverse diagonal, and is determined by the number of pixel values in each direction. The smoothness was determined by the following formula.

L is an index function having a value of {0, 1}. i and j mean the pixel values of the i-th row and j-th column in the window. L is a function that takes a value of {0, 1}, and takes 1 when the condition is true and 0 when it is false. pk, l means the pixel values of the 1kth column and the lth row in the block.
N (i, j) represents the smoothness of the block centered on the coordinates (i, j) of the target pixel of the image. The smaller the smoothness N is, the smoother it is. An example of smoothness is shown in FIG.

Next, connectivity will be described.
The connectivity is related to the property that light or dark pixels are connected in the vicinity of 4 (connected in the 4 directions of right, left and up and down). In this method, the number of bright and dark clusters is obtained from the connectivity and used for the calculation of the variability. The procedure for obtaining connectivity is described below. p [k] is an array that sequentially stores the pixel values in the block. This method is described in FIG.

Next, the degree of variability is obtained based on the obtained smoothness and connectivity.
The value of variability is a very important value in the present invention that indicates the position of the information bit to be embedded in the emblem 10. The following three points were mainly considered when calculating the variability.
(1) Whether the original pattern is smooth (2) Whether the inversion of the center pixel impairs smoothness (3) Whether the inversion of the center pixel causes a change in connectivity.

The calculation of variability is performed as follows.
(1) When the pixels of a block are dark or light and uniform, the degree of variability is set to 0 and the subsequent processing is not performed. Since this process is considered to be most abundant due to the characteristics of the input image, the speed of the process can be increased by starting the process.
(2) When the horizontal or vertical smoothness Nh, Nv is 0, 0 is assigned to the variability, and when Nh, Nv is not 0, a predetermined value is assigned to the variability as an initial value. When Nh and Nv are 0, it means that there is no change in the horizontal and vertical directions and it is very smooth. If such a block is inverted, the smoothness is remarkably impaired, so that it is omitted from the subsequent processing as a block which is not inverted.
(3) When the smoothness Nd1, Nd2 in the oblique or reverse oblique direction is 0, the degree of variability is reduced. In other cases, the threshold value T is compared with the lowest smoothness in the four directions, and the degree of variability corresponding to the difference is reduced.
(4) When the center pixel is inverted, if the vertical and horizontal smoothness does not change, the variability is increased, and if it is changed, the variability is decreased.
(5) If the number of bright and dark clusters is changed when the center pixel is inverted, the degree of variability is reduced. In the present invention, 0.5 is used as a predetermined value as an initial value of variability, 3 is used as a threshold value T, and 0.125 is used as a change in variability once. A block in which 1-bit information can be embedded when the obtained degree of variability is 0.5 or more. This operation is performed for all blocks within the effective range of the input image, and the amount of information embedded in the image is measured. Since the data can be embedded if it is larger than the encoded data amount, the padding operation is performed sequentially. In this method, it is possible to embed 1-bit information by changing the central pixel of one embeddable block. It is assumed that “0” is filled when the number of dark pixels in the block is an even number, and “1” is filled when the number of dark pixels is an odd number.

(Add frame)
S130 will be described. Adds a frame to assist in reading to an image in which information is embedded. (The frame is omitted because it was described in the description of the configuration of the Be-code 10 above.) By using this frame, the read image is slanted or distorted when reading the Be-code 10 described later. The most important feature is that the original image can be obtained without performing affine transformation or projective transformation.
First, 1 pixel bright padding 100 is arranged around the embedded image, and 2 × 2 pixels bright direction detection space 130 is arranged at each corner.

However, a dark direction detection mark 131 of one pixel is arranged at the lower right of the direction detection space 130 which is the upper left. The padding 100 is provided to prevent the dot frame 110 and the image of the emblem 10 from being connected. Although the direction detection mark 131 and the direction detection space 130 slightly invade the area of the emblem 10, the corner of the input image is rarely dark because of the characteristics of the input image. Good.

Next, around the padding 100, a dot frame 110 which is a dark 1: 1 and alternating dotted line is arranged. By the initialization process of the input image, darkness can always be placed at the corners, and light and dark are always alternated. Since one dot of the dot frame 110 serves as an index of the one-dot size of Be-code 10, it assists in reading. A bright margin 140 of one pixel is arranged around the dot frame 110. The margin is provided so that the dot frame 110 and the image outside the Be-code 10 are not connected.

(Be-code reading method)
A method of processing an image obtained by reading the Be-code 10 in which the code is recorded with a camera or the like will be described with reference to FIG.
This procedure consists of image binarization (S210), frame angle detection (S220), frame analysis (S230), Be-code restoration (S240), and reading of embedded bit information (S250).

(Binarization)
First, the input image of Be-code 10 is binarized. The same method is used for binarization as when the code was created. There are two types of binarization methods: a fixed threshold value and a dynamic threshold value. In the present invention, in order to handle an image affected by shadows or lighting, a threshold value is dynamically assigned. As a technique for dynamically assigning threshold values, a discriminant analysis method for determining threshold values by statistical processing was used.

The discriminant analysis method is obtained by classifying each group of the luminance value of the object and the luminance value of the background as one class, minimizing the variance within those classes and maximizing the variance between the two classes. This is a method of binarization using a threshold value. Further, in the present invention, in order to reduce the influence of the shadow, the image is divided into four and binarization processing is performed in each area.
As described in the case of generation of Be-code 10 for binarization, various methods have been proposed and each has its own characteristics, and is not fixed to this.

(Frame corner detection)
The explanation of FIG. 4 will be given.
The angle detection algorithm in the present invention is described below. This method can be detected quickly and accurately without performing expensive operations (processing time) such as Hough transform and trigonometric functions.

(1) X-direction search band (center and 5 pixels above and below it) passing through the center of the binarized image that is longer in length and breadth (in most mobile phones, the length is longer, so here it is assumed to be longer) ) Is scanned from the left, and the darkness at the beginning is taken as the left side of the frame. (S222)
(2) A range (2 pixels on the left and right and 10 pixels on the upper side) having the point as the lower middle point is scanned in order from the upper left. Repeat with the darkest left at the top as the next point. (S224)
(3) If there is no point on the top, set the coordinates as the upper left corner of the frame. (S226)
(4) A range (2 pixels on the left and right and 10 pixels on the bottom) with the point searched in (1) as the upper midpoint is scanned in order from the lower left. Repeat with the darkest left at the bottom as the next point.
(5) If there is no point below, the coordinate is taken as the lower left corner of the frame.
(6) Similarly, each corner is detected on the right side. (The right side is the upper right and the lower right dark is the next point.)
(7) The center of each corner is detected and used as the corner coordinates of the frame. (S228)

(Frame analysis)
The description of FIG. 5 will be given.
A line connecting the center coordinates of the corners of adjacent frames and one pixel above and below or right and left are scanned, and the coordinates of the start of all white and black dots of each line are recorded. (S232, S234)

(Restoring Be-code)
The description of FIG. 6 will be given.
In the present invention, in order to restore the photographed Be-code 10 image, the input image is restored from the analysis information of the added dot frame 110 without performing geometric processing such as projective transformation or affine transformation. If this processing can restore the same image as the digital Be-code 10 image, correct embedded data can be read. By using this method, an image in which the edge is expanded or contracted due to the blur can be read accurately.

In order to restore the captured image of the Be-code 10, first, an intersection of straight lines connecting the centers of the pth dot on the left side and the right side of the frame and the qth dot on the upper side and the lower side is obtained. The center of each dot is the midpoint of the start coordinates of that dot and the next dot. The pixel value of (p, q) of the restored image is obtained from the intersection and 8 neighborhoods of the intersection.

In the present invention, the luminance value is added and the average value is used with the intersection weight set to 1 and the neighborhood weight set to 1/8 for calculation of the pixel value. This is because the lens system blur is often Gaussian blur. Therefore, by adding a weight to neighboring pixel values, the influence of blur can be absorbed to some extent. The obtained image is binarized once again. Thereafter, the direction detection mark is detected from the positional relationship of the frames, and the restoration is performed on the Be-code 10 by matching the direction with the detected corner as the upper left.

(Experiment environment)
In the experiment, Be-code created on a PC was printed with a printer, photographed with a camera-equipped mobile phone, and the photographed image was transferred to a PC and read. In addition, the information to be embedded at the time of creating the Be-code has no meaning, and all the central pixels of the block to be embedded are inverted. The experimental environment is as follows.

The camera-equipped mobile phone used was a SHARP705H 2006 CMOS sensor with 2 million effective pixels and was photographed with a 480 × 640 close-up mode / zoom function. The printer that printed Be-code 10 is Canon's LBP-470. The experiment was conducted indoors, and the lighting condition was an indoor fluorescent lamp.

  The result of reading the embedded information of 40 bits is about 20% including errors of about 1 to 2 bits even when correction processing is not required at all. In addition, there is no problem because 100% can be recognized without any problem.

FIG. 9 is a diagram conceptually illustrating a usage form of the Be-code 10.
When a medium on which Be-code 10 is printed is read by a camera of a mobile phone, it is captured as digital data, but the data is usually a multi-valued color image.

As described above, it is binarized and converted into a binary image, the search coordinates are defined using a frame, the degree of variability is referred to, and the code is read.
The mobile phone program for executing these is configured to be downloaded from a mobile phone site or the like. Therefore, a URL for WEB access can be picked up from the read code in the mobile phone, and normal Internet access or the like can be used.

As a result, if the information embedded in the Be-code 10 is a site providing the logo, for example, the provided site can be directly browsed.
However, since service information and the like have security, it is possible to request advance registration similar to the acquisition of general site information.

  By using the present invention, code information can be embedded in a binarized image such as a simple pattern or logo, and can be easily read by a mobile phone. Moreover, even if it is shifted from the center, tilted, or slightly distorted when reading with a mobile phone, it can be read without problems.

10 Be-code whole 120 Image in which coded information is embedded
An image with embedded information, 100 padding including two colors, white and black
110 dot frame
130 Direction detection space 131 Direction detection mark 140 Margin

Claims (8)

Converting the image into a binarized image;
Embedding bit information in the binarized image;
Next , adding a frame to assist the reading to the binarized image;
A method of embedding bit information in an image characterized by comprising: The frame is a dot frame in which pixels having different brightness and darkness are alternately arranged, and the dot frame has four sides of a quadrangle, and the number of pixels on each of the four sides is an odd number. A method of embedding bit information in an image according to 1. Four direction detection spaces, which are regions in contact with four corners formed by two adjacent sides of the dot frame, are provided, and the square corner pixels are formed in one part of the direction detection spaces. The same bright or dark pixels are arranged as direction detection marks, and only the light or dark pixels different from the square corner pixels are arranged in the other three direction detection spaces. A method of embedding bit information in the image described in 2. 4. The padding which is an area in which bright or dark pixels different from the four corner pixels of the rectangle are arranged adjacent to the four sides of the dot frame. 5 . A method for embedding bit information in the image described in. A method for reading the bit information by a method of embedding bit information to the image of any one of claims 2 4 from the bit information is embedded images,
Taking the binarized image in which the bit information is embedded and binarizing;
Detecting corners of the frame;
Method of reading bit information from images embedded bit information comprising the step of reading the bit information. Detecting the corner of the frame comprises:
A region having a plurality of pixel widths including a line located in the center of the width of the image in the direction perpendicular to the scanning direction of the pixels of the binarized captured image is scanned, and faces one of the four sides of the frame. Detecting a part of two sides;
Scan from a part of the detected two sides in a direction different from the scanning direction, sequentially identify bright or dark pixels that are the same as the corner pixels of the dot frame, and determine the end of the specified pixel array Detecting as a corner of the frame;
The method of reading bit information from an image in which the bit information is embedded according to claim 5 . The step of reading the bit information includes:
Scanning from the four corner positions of the detected frame in the direction of the four sides of the frame and detecting each dot of the dot frame;
Identifying the intersection of lines connecting corresponding bright or dark dots on opposite sides as the pixel location of the image in which the bit information is embedded;
The method of reading bit information from an image in which the bit information is embedded according to claim 6 . A method of embedding bit information to the image of any one of claims 2 to 4, by a method of reading bit information from images embedded bit information described in claim 5 to 7, bit information in the image How to embed and read.